Add coordinates generation implementation

unsupervised/warp.py

@@ -72,7 +72,28 @@ def pose_vec2mat(vec):
     return transform_mat
 
 
-def projective_inverse_warp(target_img, source_img, depth, pose, intrinsics):
+def image_coordinate(batch, height, width):
+    """
+    Construct a tensor for the given height/width with elements the homogenous coordinates for the pixel
+    :param batch: Number of images in a batch
+    :param height: Height of image
+    :param width: Width of image
+    :return: Tensor of shape (B, height, width, 3), homogenous coordinates for an image.
+        Coordinates are in order [x, y, 1]
+    """
+    x_coords = tf.range(width)
+    y_coords = tf.range(height)
+
+    x_mesh, y_mesh = tf.meshgrid(x_coords, y_coords)
+
+    ones_mesh = tf.cast(tf.ones([height, width]), tf.int32)
+
+    stacked = tf.stack([x_mesh, y_mesh, ones_mesh], axis=2)
+
+    return tf.repeat(tf.expand_dims(stacked, axis=0), batch, axis=0)
+
+
+def projective_inverse_warp(target_img, source_img, depth, pose, intrinsics, coordinates):
     """
     Calculate the reprojected image from the source to the target, based on the given depth, pose and intrinsics
 
@@ -88,12 +109,14 @@ def projective_inverse_warp(target_img, source_img, depth, pose, intrinsics):
     :param depth: Tensor, (batch, height, width, 1)
     :param pose: (batch, 6)
     :param intrinsics: (batch, 3, 3)
+    :param coordinates: (batch, height, width, 3) - coordinates for the image. Pass this in so it doesn't need to be
+        calculated on every warp step
     :return: The source image reprojected to the target
     """
     # Convert pose vector (output of pose net) to pose matrix (4x4)
 
     # Convert intrinsics matrix (3x3) to (4x4) so it can be multiplied by the pose net
-    intrinsics_4x4 =
+    # intrinsics_4x4 =
 
     # Calculate inverse of the 4x4 intrinsics matrix
     tf.linalg.inv()